Closed Bug 1679700 Opened 4 years ago Closed 4 years ago

Magenta overlay on videos (macOS Intel Gen9 Skylake)

Categories

(Core :: Graphics, defect, P3)

Product:
Core
Component:
Graphics
Type:
defect

Tracking

()

Status:
RESOLVED FIXED
Milestone:
85 Branch
Tracking Flags:
Tracking Status
firefox-esr78 --- unaffected
firefox83 --- unaffected
firefox84 + fixed
firefox85 + fixed

People

(Reporter: flod, Assigned: mstange)

References

(Blocks 1 open bug, Regression)

Details

(Keywords: perf-alert, regression)

Attachments

(4 files)

immagine.png
298.04 KB, image/png
Details
about:support (work)
52.62 KB, text/plain
Details
aboutsupport.txt
25.47 KB, text/plain
Details
Bug 1679700 - Cast int parameter in get_yuv_offset_vector to float, to work around a shader compilation bug with integer equality comparisons on macOS Intel. r=jrmuizel
47 bytes, text/x-phabricator-request
RyanVM
: approval-mozilla-beta+
Details | Review
Attached image immagine.png —

I'm seeing a magenta overlay over videos (in Twitter, but also previews on Netflix).

Firefox 85 (20201129092505), macOS 10.15.7 (19H2), macbook pro with Intel Iris Graphics 550.
I assumed it started after the last update.

Bugbug thinks this bug should belong to this component, but please revert this change in case of error.

Component: Untriaged → Graphics
Product: Firefox → Core

Updated from 20201126212448 (first to 20201127155321, then 20201129212213) on another macbook 13 with Intel Iris Plus Graphics 650, same problem.

tracking-firefox85: --- → ?

https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=ff2a53ac9e8ba797b2c4c88dc97009cc03caed15&tochange=b2d438ba2bdd2f2098ce1d4b3be1fd66c7eecf46

It seems to be bug 1678924, not sure if it's a dupe of bug 1678903 at this point (it doesn't look like, given how long that's been around).

Regressed by: 1678924
Has Regression Range: --- → yes

Set release status flags based on info from the regressing bug 1678924

status-firefox83: --- → unaffected
status-firefox84: --- → unaffected
status-firefox85: --- → affected
status-firefox-esr78: --- → unaffected
Blocks: 1679764

This might be a regression on Gen9 Skylake from the Gen6 fix.

Flags: needinfo?(jmuizelaar)
Priority: -- → P3

Flod, mind posting your about:support text?

Flags: needinfo?(francesco.lodolo)
Summary: Magenta overlay on videos → Magenta overlay on videos (Intel Gen9 Skylake)
Severity: -- → S2
Attached file about:support (work) —

This is from the work computer (Iris 650, MacBook Pro (13-inch, 2017, Four Thunderbolt 3 Ports))

Flags: needinfo?(francesco.lodolo)

Also, worth noting that this doesn't happen on YouTube, for example. So far, I've seen it in on Twitter and Netflix (only for the previews when hovering an element). If it helps, I can provide also the about:support from the other machine.

Attached file aboutsupport.txt —

Here is another about:support info from my MacBook Pro 2018 with Intel UHD Graphics 630 which is also affected.

I can't reproduce this on a UHD Graphics 630 running 10.15.6

Flags: needinfo?(jmuizelaar)

I can see the same problem with Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:85.0) Gecko/20100101 Firefox/85.0 ID:20201130093031 and Intel Inc. -- Intel(R) Iris(TM) Plus Graphics 655.

For confirmation I run mozregression and found bug 1678924 being the culprit here:

 6:06.70 INFO: Narrowed integration regression window from [ff2a53ac, bc148cb7] (3 builds) to [ff2a53ac, b2d438ba] (2 builds) (~1 steps left)
 6:06.70 INFO: No more integration revisions, bisection finished.
 6:06.70 INFO: Last good revision: ff2a53ac9e8ba797b2c4c88dc97009cc03caed15
 6:06.70 INFO: First bad revision: b2d438ba2bdd2f2098ce1d4b3be1fd66c7eecf46
 6:06.70 INFO: Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=ff2a53ac9e8ba797b2c4c88dc97009cc03caed15&tochange=b2d438ba2bdd2f2098ce1d4b3be1fd66c7eecf46

I'm also seeing this on my 16 inch 2019 Macbook Pro on some but not all Twitter videos and a few other places. Here's an example where I see it, because it looks like one hasn't been posted yet: https://twitter.com/catsdotexe/status/1333268668755668992 (if you click play, it goes magenta)

(In reply to Jeff Muizelaar [:jrmuizel] from comment #13)

I started to try pushes:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=7c6fce7ae9870de98c94bffd4f07570fe1f47995

This build seems to work correctly for me (at least, Twitter does).

https://treeherder.mozilla.org/#/jobs?repo=try&revision=c7f908a948867267b30e7633c6742fae8dac6255

This still has the issue.

Used this tweet as a test https://twitter.com/firefox/status/1333436877861433347

Summary: Magenta overlay on videos (Intel Gen9 Skylake) → Magenta overlay on videos (macOS Intel Gen9 Skylake)

We've encountered similar bugs before, e.g.
https://github.com/servo/webrender/wiki/Driver-issues#3540---mac-glsl-compiler-bug-with-integer-comparison-take-two

We don't want to change this if to a switch because glsl-opt compiles it
down to an if statement anyway, just a more complicated one (which was triggering
other bugs on other platforms, see bug 1678924).

Assignee: nobody → mstange.moz
Status: NEW → ASSIGNED
Pushed by mstange@themasta.com:
https://hg.mozilla.org/integration/autoland/rev/fdf86dfacb7c
Cast int parameter in get_yuv_offset_vector to float, to work around a shader compilation bug with integer equality comparisons on macOS Intel. r=jrmuizel
Pushed by mstange@themasta.com:
https://hg.mozilla.org/integration/autoland/rev/5d4b227d3b5c
Cast int parameter in get_yuv_offset_vector to float, to work around a shader compilation bug with integer equality comparisons on macOS Intel. r=jrmuizel

For the record: This happens to me on Google Meet.

Blocks: meet
Flags: needinfo?(mstange.moz)

https://hg.mozilla.org/mozilla-central/rev/5d4b227d3b5c

Status: ASSIGNED → RESOLVED
Closed: 4 years ago
status-firefox85: affected → fixed
Resolution: --- → FIXED
Target Milestone: --- → 85 Branch

Confirming the fix for Nightly 85.0a1 (2020-12-01). Google Meet works again. Thank you!

Comment on attachment 9190342 [details]
Bug 1679700 - Cast int parameter in get_yuv_offset_vector to float, to work around a shader compilation bug with integer equality comparisons on macOS Intel. r=jrmuizel

Beta/Release Uplift Approval Request

  • User impact if declined: Magenta video on macOS
  • Is this code covered by automated tests?: No
  • Has the fix been verified in Nightly?: Yes
  • Needs manual test from QE?: No
  • If yes, steps to reproduce:
  • List of other uplifts needed: Bug 1678924
  • Risk to taking this patch: Low
  • Why is the change risky/not risky? (and alternatives if risky): This only risk here is other shader compiler bugs. That's not very likely.
  • String changes made/needed:
Attachment #9190342 - Flags: approval-mozilla-beta?

Comment on attachment 9190342 [details]
Bug 1679700 - Cast int parameter in get_yuv_offset_vector to float, to work around a shader compilation bug with integer equality comparisons on macOS Intel. r=jrmuizel

Approved for 84.0b7.

Attachment #9190342 - Flags: approval-mozilla-beta? → approval-mozilla-beta+

== Change summary for alert #28015 (as of Tue, 08 Dec 2020 12:35:54 GMT) ==

Improvements:

Ratio Suite Test Platform Options Absolute values (old vs new)
23% espn LastVisualChange android-hw-p2-8-0-android-aarch64-shippable nocondprof warm 6,970.67 -> 5,357.33
22% espn LastVisualChange android-hw-p2-8-0-android-aarch64-shippable nocondprof warm webrender 6,898.88 -> 5,373.75
6% espn SpeedIndex android-hw-p2-8-0-android-aarch64-shippable nocondprof warm webrender 1,639.92 -> 1,547.25
4% espn SpeedIndex android-hw-p2-8-0-android-aarch64-shippable nocondprof warm webrender 1,624.54 -> 1,555.92

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=28015

Keywords: perf-alert

Isn't that a Mac specific issue? How could this have been caused performance improvements on Android? The graph shows lot of jumps up and down, and since recently isn't that stable anymore. I get the feeling that a different patch actually caused that.

Flags: needinfo?(fstrugariu)

Did some rebuilds will come back with a result

I agree with Henrik that this bug is not the cause of the improvement.
as there are some issues running the test in the commits around this bug and that the grephs came back to the before value I ill mark this as a infrastructure issue.
Please ignore the previous alert comment

Flags: needinfo?(fstrugariu)
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: